Sei sulla pagina 1di 9

The Enumeration of Preimages and Gardens-of-Eden in Sequential Cellular Automata

Yan Deqin Hou Qinghu

Research Center for Combinatorics, Nankai University, Tianjin 300071, China

The enumeration of preimages in sequential cellular automata is ap- proached. Two methods are given to enumerate the preimages of a rule. Preimage enumeration is simplified by using transform matrix expres- sions. The concept of factor graphs is presented in the study of De Bruijn graphs. A criterion for a rule having gardens-of-Eden is given.

1. Introduction

Cellular automata (CA) are discrete dynamic systems, of simple con- struction but varied behavior. Studying preimages of CA is important for learning their properties [2, 3]. Gardens-of-Eden (GOE) are se- quences with no preimages and are related to the preimages of a rule in [1, 2, 4]. Recently in [5], Chris Barrett, William Y. C. Chen, and Christian Reidys made an approach on the GOE in sequential cellular automata (SCA), and proposed a question: What is a sufficient and necessary condition of a rule having no gardens-of-Eden (i.e., to be non- GOE)? For the enumeration of preimages in parallel CA, E. Jen gave a recurrence relation formula in [1]. But in SCA there is no formula to enumerate the preimages. This paper presents a preimages enumeration formula for a rule in SCA, and gives two methods for enumerating the preimages of a rule. The enumeration is greatly simplified by using a transform matrix ex- pression, especially when r and k are bigger than 1 and 2. The concept of factor graphs is presented in the study of sequential De Bruijn graphs. A criterion for a rule being nonGOE is given. That is, a certain kind of special structure of rule matrix is the sufficient and necessary condition for a rule to be nonGOE.

Electronic mail address: hqh@public1.tpt.tj.cn.

Complex Systems, 12 (2000) 83–91; c 2000 Complex Systems Publications, Inc.

84

Y. Deqin and H. Qinghu

2. Preliminaries

The general form of SCA is

= f ( x

x

t

i

+ 1

t

+

1

ir ,

+

t

, x

i

1

1

t

, x , x

i

t

i+ 1 ,

, x

t

i+ r )

where x denotes the value of site i at time t . The values are restricted

, k 1} , f represents the “rule” defining

to the finite field F k = { 0, 1,

the automaton and r is a nonnegative integer specifying the radius of the rule.

Just as in parallel CA, the rule number for primary CA (r = 1,

k = 2) is defined as the decimal value of the term’s binary expression:

a

,

111} . i is the decimal value of xyz . For example, according to a rule

with f (111) = f (110) = f (101) = f (100) = 0, f (011) = f (010) = f (001) = f (000) = 1, the rule number is 15, and is called rule 15.

t

i

7 a 6

a 0 , where a i = f ( xyz )( i = 0,

, 7). xyz { 000, 001,

Definition 1. A k 2r × k 2r matrix, with rows and columns labeled by the sequences of length 2r on F k , is called the rule matrix of rule f

if P i,j = x 0 = f ( m 0 , m 1 ,

m

m 2r 1 x; and P i,j = 0, otherwise. Also, let the

m 2r 1 , j =

, m 2r 1 , x), for i = m 0 m 1

1 m 2

m r 1 x 0 m r + 1

vertices be the sequences of length 2r and draw an arc from i to j with the label x 0 for all i, j as above. Then the digraph is called a sequential

De Bruijn graph of the rule.

When r = 1, k = 2, P is a 4 × 4 matrix. The row and column labels

are 00, 01, 10, and 11. The element of row (y 1, y 2) and column ( y 0, y )

is y 0 = f ( y 1, y 2, y ).

If y 0 = 0, the element is denoted by α, if y 0 = 1, it

is denoted by β , other elements are 0. For example, the rule matrix of

rule 15 is

0

0

α

α

0

0

α

α

β

β

0

0

β

β

0

0

.

In fact, this rule matrix is the depiction of a De Bruijn graph. The De Bruijn graph in CA and in SCA of rule 121 are shown by D1 and D2 in Figure 1 respectively.

3. Preimage enumeration in sequential cellular automata

For an arbitrary integer 0 m k 2r 1, with

m =

2r 1

i= 0

m i k 2r i1 ,

Complex Systems, 12 (2000) 83–91

The Enumeration of Preimages
The Enumeration of Preimages

85

Figure 1. The De Bruijn graph of rule 121.

m 2r 1 the symbol corresponding to its k-ary

representation. The symbols M thus range over all possible blocks of length 2r.

s 1 , the number of its preimages is

denote by M = m 0 m 1

Given a sequence S = s n s n 1

denoted by N ( S ). Let L M j be the preimage of the sequence s j beginning with M, from [1] we have

s 1

N ( S ) =

k 2r 1

m= 0

L

n

M ,

where

L

j

M =

i

L

j

M

i

1

I j ( x i ),

(1)

and

I j ( x) =

1, x = x j ,

0, otherwise.

Definition 2. Let L j be the column vector consisting of L M j , then equa- tion (1) can be written as

(2)

We call P j the transform matrix.

In fact, the rule matrix has the same structure as the transform matrix.

a is the sequence produced by a

Theorem 1. On F k , suppose S = aa rule with radius r, then

L j = P j L j 1 .

L n = A n E,

where E is the k 2r -dimension column vector and A is the transform matrix of the rule on S .

Complex Systems, 12 (2000) 83–91

86

Y. Deqin and H. Qinghu

Proof . From the structure of the transform matrix, we know P 1 = P 2 = ··· = P j , and by equation (2) we get L n = A n E.

P j , and by equation (2) we get L n = A n E .

Theorem 2. On F k , for any n element sequence S , we separate S into blocks such that in one block all of the elements are the same. Let v i be the value of the elements in block i and u i be the length of the block. We have

(3)

L n = A u t

v

t

A u 1 E,

v

1

where v i F k , t , u i are integers, E is the k 2r element column unit vector (all elements are 1), and A u i is the transform matrix on the elements of

v i ( i = 1, 2,

Proof . For a sequence S , the separation method means that if s i = s i 1 =

, there are v 1 = s 1 , u 1 = i;

v 2 = s i+ 1 , u 2 = j

··· = s 1 , s i+ j+ 1 = s i+ j = ··· = s i+ 1 = s i

v

i

, t ).

From Theorem 1 we have

Then

L u 1 = L u 2 =

L u n = A u n · L u n 1 .

A u 1 E, A u 2 ·

v

v

1

2

v

n

L u 1 ,

,

L n = A u t

v

t

A u 1 E.

v

1

u 1 , , L n = A u t v t A u 1 E.

There are two ways to enumerate the preimages in SCA. One is to apply the method used in [1]; the other is to use Theorem 2, which we call the transform matrix method. In the following we take r = 1, k = 2 and rule 12 as an example for enumerating the preimages by both methods. Let S be an arbitrary sequence. From right to left, divide S into blocks of consecutive 0s and 1s, and let

a i = number of consecutive 0s in the ith block, b i = number of consecutive 1s in the ith block.

For example, with S = 00101001100100, set a 1 = a 2 = a 3 = 2, a 4 = 1, a 5 = 2 and b 1 = 1, b 2 = 2, b 3 = b 4 = 1.

b 1 a 1 . On

the block of 0s the transform matrix is A , and on the block of 1s the

transform matrix is B. By using the first method, we have

Suppose there is a sequence S with the blocks of b t a t

L

L

j

00

j

10

= ( L j 1

00

= ( L j 1

00

+

+

L

L

j

01

j

01

1

1

) I j (0), ) I j (0),

L

L

j

01

j

11

= ( L j 1

10

= ( L j 1

00

+

+

L

L

j

11

j

01

1

1

)

)

I j (1), I j (0),

Complex Systems, 12 (2000) 83–91

The Enumeration of Preimages

87

and

Then

U j

00

= U j1 + U j1

00

01

U 01 j = 0,

U j

= U j1 + U j1

00

01

10

U j

11

= U j1 + U j1

00

01

V 00 j = 0,

V j

= V j1 + V j1

10

11

01

V j

10

=

V j

11 = 0.

,

,

,

,

1

U

U a

00

1

00

= U 10 = U

= U 10 a = U a

1

1

1

11

1

11

U 01 a = 0,

V

1

b

1

00

= V 10 = V

1

b

b

11

1

= 2, U 01 = 0,

= 2,

1

= 0.

If b 1 > 1, there is V 01 = 0. If b 1 = 1, there is V

1

1

b

b

01 = U 10 a + U a

1

11 1 = 4.

1

For the first a 2 0s there is U 00 a = ··· = U 00 . After a 2 0s there is a 1,

2

according to the rule, U

00 = V 01 = 4, namely U 00 a = 4. So we have

1

2

1

b

U

a

10 = U a

2

11 = 0,

2

U 01 a = 0.

2

From the recurrence relation of V 01 j , we know b i = 1, ( i = 1, 2, According to the rule, there is

).

V

i

i

b

01 = U 10 a + U 11 a i

b

i

b

i

00 = V 10 = V 11 = 0

V

b

i

so

U

Then

a 10 = U a

i

i

11 = V

b i 1

01

.

V 01 = 2U 11 a = 2V

i

i

b

b i 1

01

b

01

= 2 i 1 V

1

N ( S ) =

i,j{0,1}

b

ij

V

t

b

01

= V

t

= 2 t + 1 .

= 2 i+ 1 ,

By using the transform matrix method we have

A =

0000

0011

0000

0011

,

B =

1100

0000

1100

0000

.

Complex Systems, 12 (2000) 83–91

88

Y. Deqin and H. Qinghu

88 Y. Deqin and H. Qinghu Figure 2. The factor graphs of the De Bruijn graph

Figure 2. The factor graphs of the De Bruijn graph of a nonGOE rule.

It is easy to see that B i = B and A j = 0 ( i = 1, 2, L n = C t E, where

C = AB =

   ,

0

2

0

0

0

2

0

0

0

0

0

0

0

0

0

0

 

and hence

L n =

2

0

t + 1

0

0

.

So we have

N ( S ) = 2 t + 1 .

; j 2). We have

4. The no gardens-of-Eden matrix expression

s 1 such that N ( S ) = 0, we say the rule

has GOE, otherwise it is nonGOE.

1 as 0 and 1 . For any sequence S , if

we find a path in the De Bruijn graph of a rule, and there is a bijective

map between S and the consecutive labels of the path, it is denoted as { 0, 1} . Since the preimages of the sequence S are the combinations of the vertices between which the labels hold consecutive maps to the elements of S , the rule is nonGOE.

Given a rule, if there is S = s n

We denote 0

0 and 1

Definition 3. We call graphs T1 and T2 in Figure 2 the factor graphs of a nonGOE rule.

Proposition 1.

or 1 .

In graphs T1 and T2, one can find the path labeled 0

Complex Systems, 12 (2000) 83–91

The Enumeration of Preimages
The Enumeration of Preimages

89

The Enumeration of Preimages 89 Figure 3. The graphs delivered from T1. Proof . In graph

Figure 3. The graphs delivered from T1.

Proof . In graph T1, from A1 to B1 or A1 via C1 to B1 is the path. In graph T2, from A2 to B2 or A2 to C2 via several cycles made of B2 and C2 to one of B2 and C2 is the path.

Graphs T1 and T2 can be changed to graphs T3 and T4, T6 is another form of T1. Combinations of the factor graphs can produce the { 0, 1} path shown in Figure 3.

can produce the { 0, 1 } ∗ path shown in Figure 3. Theorem 3. To

Theorem 3. To obtain the graph containing the { 0, 1} path by using graphs T1, T2, T3, and T4, the possible combinations are T1 + T2, T3 + T3, and T4 + T4.

Proof . Label the arcs or ring with 0 in graph T1, and 1 in graph T2. Let A1 connect B2 and B1 connect A2, then we can find the { 0, 1} path in the new graph. Similarly, we can find the { 0, 1} path in the new graph of T3 + T3 and T4 + T4. Now we prove it is impossible to get the { 0, 1} path from the other combinations. T1 + T3: The only combination is vertex B3 connects A3. But, according to the rules in SCA, the label of the cycle at vertex A3 must be the same as the label in T1. So it is impossible. T1 + T4: The only combinations are vertex B1 connects A4 and A1 connects B4. If so, there must be three arcs to get out from vertex B4, and it is impossible in SCA. For the same reason, T2 + T3, T2 + T4, and T3 + T4 are not qualified.

In SCA, according to the definition of the De Bruijn graph, there are four cases of combinations of T1 (or T6)+ T2.

there are four cases of combinations of T1 (or T6) + T2. (a) B1 connects A2,

(a)

B1 connects A2, A1 connects B2.

(b)

C1 connects A2, A1 connects C2.

(c)

B6 connects A2, Q connects B2, P connects C2.

(d)

B6 connects A2, P connects B2, Q connects C2.

At first, we label all arcs and rings in graphs T1 and T6 with 0s, that is, let graphs T1 and T6 produce the 0 path. Then graph T2 must

Complex Systems, 12 (2000) 83–91

90

Y. Deqin and H. Qinghu

produce the 1 path, which implies that all arcs in factor graph T2 are labeled 1. For cases (a) and (b), the vertex B1,C1 must be 00,01 or 01,00. The vertex A1,C2 can be any of 10,11. The rule matrices must be of the form

P 1 =

P 3 =

α

0

α

0

α

0

0

α

α

0

α

0

α

0

0

α

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

,

,

P 2 =

P 4 =

0

α

α

0

0

α

0

α

0

α

α

0

0

α

0

α

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

,

.

In the matrix “.” can be 0 or β . (This explanation is validated in the following.) For the same reason, in cases (c) and (d) the vertex B1,C1 must be 00,01 or 01,00. The corresponding matrices are

P 5 =

P 7 =

α

0

0

α

α

0

α

0

α

0

α

0

α

0

0

α

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

,

,

P 6 =

P 8 =

0

α

α

0

0

α

0

α

0

α

0

α

0

α

α

0

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

,

.

On the other hand, we let factor graph T2 produce the 0 path, and T1 and T6 produce the 1 path. The matrices with the structure of the combined graph using cases (a), (b), (c), and (d) are

P 9 =

0

α

α

0

α

0

α

0

.

.

.

.

.

.

.

.

,

P 10 =

0

α

0

α

α

0

0

α

.

.

.

.

.

.

.

.

.

Now we consider graph T5. T5 is the self-combination of T3 or T4, two T5s constitute a De Bruijn graph. When vertex A5 is 00 or 01, vertex B5 must be 10 or 11. Then the four vertices of the De Bruijn graph are 00,10 (or 00,11) and 01,10 (or 01,11). The corresponding matrices are

P 11 =

α

.

.

α

.

.

.

.

.

.

.

.

Complex Systems, 12 (2000) 83–91

β

.

.

β

,

P 12 =

.

.

.

.

.

α

α

.

.

β

β

.

.

.

.

.

,

The Enumeration of Preimages

P 13 =

.

.

,

.

α

.

.

α

β

β

.

.

.

 

.

P 14 =

  α

   .

 

β

β

.

.

.

.

α

91

It can be seen that a rule matrix determines the property of being nonGOE. However, any { 0, 1} path must include the factor graphs; and, in order to get the { 0, 1} path, we have discussed all possible combinations. So we get the conclusion in Theorem 4.

Theorem 4. A rule is nonGOE if and only if the structure of its rule

matrix is the same as one of P 1 ,

, P 14 .

Example 1. The rule matrix of rule 150 is

   .

α

0

0

α

0

α

α

0

0

β

β

0

0

β

 

β

0

Its structure is the same as P 11 and P 12 . So rule 150 is nonGOE. In fact, there are two factor graphs of T5 in its De Bruijn graph.

Acknowledgments

We are grateful to our advisor, professor William Y. C. Chen, for his help.

References

[1] E. Jen, “Enumeration of Preimages in Cellular Automata,” Complex Sys- tems, 3 (1989) 421.

[2] E. Jen, “Scaling of Preimages in Cellular Automata,” Complex Systems, 1 (1987) 1045.

[3] S. Wolfram, Theory and Applications of Cellular Automata (World Scien- tific Press, Singapore, 1986).

[4]

S. Amoroso and G. Cooper, “The Garden-of-Eden Theorem for Finite Con-

figurations,” Proceedings of the American Mathematical Society, 26 (1970)

158.

[5]

Chris Barrett, William Y. C. Chen, and Christian Reidys, “A Combinatorial Study of Parallel and Sequential Cellular Automata” (preprint).

Complex Systems, 12 (2000) 83–91